Moisture Classification of Gas–Solid Two-Phase Flow Based on a Pseudo-Siamese Neural Network

For gas–solid two-phase flow, the moisture of solid particles affects the change of flow movement and flow field greatly, thereby impacting industrial output. Different measurement methods have been employed to study the effects of moisture under different conditions, and these methods often yield complementary results. Therefore, a classification method for two-phase flow, based on a pseudo-Siamese neural network (pSNN), is proposed. Using a 12-electrode electrical capacitance tomography (ECT) sensor and a charge-coupled device (CCD) camera, we conduct dynamic experiments under five different moisture conditions to collect ECT data and image data. The observation directions of these data are perpendicular to each other. Afterward, we develop a two-layer long short-term memory (LSTM) network and a residual neural network (ResNet) to train on ECT and image data, respectively. Additionally, we add a granularity selection experiment to the ECT data subnetwork training. By feature concatenation, the overall model can forecast moisture more accurately than a single measurement method, with an accuracy of 99.3%, and it also acquires 3-D dynamic information on gas–solid two-phase flow through data fusion.